Vermont lawmakers gave final approval to a bill that would place some of the state’s first explicit limits on how artificial intelligence can be used in health care, drawing a line between administrative assistance and clinical decision-making in mental health treatment.
The legislation, H.816, was finalized in conference committee last week and approved by both chambers Tuesday after months of discussion over whether evolving AI tools should have a place in therapy, diagnosis or treatment planning. The bill next heads to Gov. Phil Scott for his consideration.
If signed into law, the measure would prohibit licensed mental health providers from relying on AI systems to make therapeutic decisions or provide mental health treatment independently. Supporters say the bill would prevent large language models from effectively operating as unlicensed therapists while still allowing clinicians to use software for administrative work such as note taking, scheduling and transcription.
Rep. Daisy Berbeco (D-Winooski), who introduced the bill, previously told lawmakers the aim is to strike a balance between innovation and patient safety. Nationwide, policymakers’ concerns about AI use in therapy have intensified as reports have surfaced of people — particularly children and teenagers — receiving harmful mental health advice, including encouragement of self-harm or suicide, from chatbots. Lawmakers ultimately decided not to address chatbots explicitly in H.816 and instead focus on artificial intelligence more broadly.
The bill’s passage marks one of the legislature’s most concrete attempts to regulate use of artificial intelligence in health care. Vermont is among a growing number of states moving to address how generative AI is entering health care amid a lack of national standards.
Rep. Alyssa Black (D-Essex), chair of the House Health Care Committee, said on Tuesday that the bill allows practitioners to use AI while affirming that care must be delivered by licensed providers.
AI has been spreading across health care. Hospital administrators and providers in Vermont already use artificial intelligence, from software that transcribes patient visits and drafts clinical notes to diagnostic tools that help doctors quickly detect strokes, fractures and other medical abnormalities. Those technologies would not be affected by the bill.
But lawmakers and mental health care advocates argued that therapy presents unique risks if combined with artificial intelligence. They warned mental health treatment platforms built to mimic empathy or provide conversational support could blur legal and ethical boundaries in care settings.
Under the bill, using AI in therapeutic decision-making would constitute “unprofessional conduct,” potentially exposing providers to consumer protection enforcement or disciplinary action, such as denial of a professional license. The legislation also directs Vermont’s Office of Professional Regulation to develop recommendations for broader oversight of the technology’s use by mental health professionals before next year.
The conference committee voted unanimously to support its version of the bill, which passed both the Senate and House on voice vote without opposition. If signed by Gov. Phil Scott, the law’s provisions would take effect immediately.